ebook img

Regression Analysis Under A Priori Parameter Restrictions PDF

244 Pages·2012·2.06 MB·English
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Regression Analysis Under A Priori Parameter Restrictions

Springer Optimization and Its Applications VOLUME54 ManagingEditor PanosM.Pardalos(UniversityofFlorida) Editor–CombinatorialOptimization Ding-ZhuDu(UniversityofTexasatDallas) AdvisoryBoard J.Birge(UniversityofChicago) C.A.Floudas(PrincetonUniversity) F.Giannessi(UniversityofPisa) H.D.Sherali(VirginiaPolytechnicandStateUniversity) T.Terlaky(McMasterUniversity) Y.Ye(StanfordUniversity) AimsandScope Optimization has been expanding in all directions at an astonishing rate during the last few decades. New algorithmic and theoretical techniques havebeendeveloped,thediffusionintootherdisciplineshasproceededata rapidpace,andourknowledgeofallaspectsofthefieldhasgrownevenmore profound.Atthe sametime, oneofthe moststriking trendsin optimization is the constantly increasing emphasis on the interdisciplinary nature of the field.Optimizationhasbeenabasictoolinallareasofappliedmathematics, engineering,medicine,economics,andothersciences. The series Springer Optimization and Its Applications publishes under- graduate and graduate textbooks, monographs and state-of-the-art exposi- tory work that focus on algorithms for solving optimization problems and alsostudyapplicationsinvolvingsuchproblems.Someofthetopicscovered include nonlinear optimization (convex and nonconvex), network flow problems, stochastic optimization, optimal control, discrete optimization, multi-objectiveprogramming,descriptionofsoftwarepackages,approxima- tiontechniquesandheuristicapproaches. Forfurthervolumes: http://www.springer.com/series/7393 Pavel S. Knopov • Arnold S. Korkhin Regression Analysis Under A Priori Parameter Restrictions 123 PavelS.Knopov ArnoldS.Korkhin DepartmentofMathematicalMethods DepartmentofEconomicalCybernetics ofOperationResearch andInformationTechnology V.M.GlushkovInstituteofCybernetics NationalMiningUniversity NationalAcademyofScienceofUkraine 49005Dnepropetrovsk 03187Kiev Ukraine Ukraine [email protected] [email protected] ISSN1931-6828 ISBN978-1-4614-0573-3 e-ISBN978-1-4614-0574-0 DOI10.1007/978-1-4614-0574-0 SpringerNewYorkDordrechtHeidelbergLondon LibraryofCongressControlNumber:2011935145 ©SpringerScience+BusinessMedia,LLC2012 Allrightsreserved.Thisworkmaynotbetranslatedorcopiedinwholeorinpartwithoutthewritten permission of the publisher (Springer Science+Business Media, LLC, 233 Spring Street, New York, NY10013, USA),except forbrief excerpts inconnection with reviews orscholarly analysis. Usein connectionwithanyformofinformationstorageandretrieval,electronicadaptation,computersoftware, orbysimilarordissimilarmethodologynowknownorhereafterdevelopedisforbidden. Theuseinthispublicationoftradenames,trademarks,servicemarks,andsimilarterms,eveniftheyare notidentifiedassuch,isnottobetakenasanexpressionofopinionastowhetherornottheyaresubject toproprietaryrights. Printedonacid-freepaper SpringerispartofSpringerScience+BusinessMedia(www.springer.com) Preface Regressionanalysishasquitealonghistory.Itisconventionaltothinkthatitgoes back to the works of Gauss on approximation of experimental data. Nowadays, regression analysis represents a separate scientific branch, which is based on optimizationtheoryandmathematicalstatistics.Formally,thereexisttwobranches ofregressionanalysis:theoreticalandapplied. Up to recent time, developments in regression analysis were based on the hypothesisthatthedomainofregressionparametershasnorestrictions.Divergence from that approach came later on when equality constraints were taken into account, which allowed use of some a priori information about the regression model.Methodsof constructingthe regressionwith equality constraintswere first investigatedinRao(1965)andBard(1974). Usage of inequality constraints in a regression model gives much more pos- sibilities to utilize available a priori information. Moreover, the representation of theadmissibledomainofparametersintheformofinequalityconstraintsnaturally includesthecaseswhenconstraintsaregivenasequalities. Propertiesoftheregressionwithinequalityconstraintsareinvestigatedinmany papers,inparticular,inZellner(1971),Liew(1976),NagarajandFuller(1991)and ThomsonandSchmidt(1982),wheresomeparticularcasesareconsidered.Detailed qualitativeanalysis of the propertiesof estimates in case of linear regressionwith linearconstraintsisgiveninthemonograph(Malinvaud1969,Section9.8). Asymptotic properties of the estimates of regression parameters in regression withfinitenumberofparametersundersomeknownaprioriinformationarestudied in Dupacovaand Wets (1986),Knopov(1997a–c),Korkhin(1985),Wang (1996), etc. We note that the results obtained in Korkhin (1985) and Wang (1996) under different initial assumptions, almost coincide. There are many results concerning practical implementation of regression models with inequality constraints, for example, Liew (1976), Rezk (1996) and McDonald (1999), Thomson (1982), Thomson and Schmidt (1982). This problem was also studied in Gross (2003, Subsection3.3.2). Inthismonograph,wepresentinfulldetailtheresultsonestimationofunknown parametersin regressionmodelsundera prioriinformation,describedin the form v vi Preface ofinequalityconstraints.Thebookcoverstheproblemofestimationofregression parametersas well as the problemof accuracyof such estimation. Both problems are studied is cases of linear and nonlinear regressions. Moreover, we investigate the applicability of regression with constraints to problems of point and interval prediction. Thebookisorganizedasfollows. InChapter1,weconsidermethodsofcalculationofparameterestimatesinlinear and nonlinear regression with constraints. In this chapter we describe methods of solving optimization problems which take into account the specification of regressionanalysis. Chapter2isdevotedtoasymptoticpropertiesofregressionparametersestimates inlinearandnonlinearregression.Bothcasesofequalityandinequalityconstraints areconsidered. In Chapter 3, we consider various generalizations of the estimation problem by the least squaresmethod in nonlinearregressionwith inequality constraintson parameters.Inparticular,wediscusstheresultsconcerningrobustHuberestimates andregressorswhicharecontinuousfunctionsoftime. Chapter 4 is devoted to the problem of accuracy estimation in (linear and nonlinear)regression,whenparametersareestimatedbymeansoftheleastsquares method. InChapter5,wediscuss/considerstatisticalpropertiesofestimatesofparameters innonlinearregression,whichareobtainedoneachiterationofthesolutiontothe estimationproblem.HereweusealgorithmsdescribedinChap.1.Obtainedresults mightbeusefulinpracticalimplementationofregressionanalysis. Chapter 6 is devoted to problemsof predictionby linear regression with linear constraints. Kiev,Ukraine PavelS.Knopov Dnepropetrovsk,Ukraine ArnoldS.Korkhin Acknowledgments Weareverygratefultothescientificeditorofthisbook,ProfessorPanosPardalos, seniorpublishingeditor,ElizabethLoew,andtotheassociateeditorinmathematics, Nathan Brothers, for their helpful support and collaboration in preparation of the manuscript. We thank our colleagues from V.M. Glushkov Institute of Cybernetics of National Academy of Science of Ukraine for many helpful discussions on the problemsandresultsdescribedandpresentedinthisbook. We thank our colleagues L. Belyavina, L. Vovk, V. Knopova, Yu. Kolesnik, E.Odinzova,forinvaluablehelpduringthepreparationofourbookforpublication. vii Contents 1 EstimationofRegressionModelParameters withSpecificConstraints.................................................... 1 1.1 Estimationofthe ParametersofaLinearRegression withInequalityConstraints............................................. 2 1.1.1 MethodofEstimatingtheSolutionto(1.7) ................... 2 1.1.2 AlgorithmofFindingtheSolutionto(1.9).................... 5 1.1.3 SpecialCaseoftheProblem(1.7) ............................. 6 1.2 Estimation of Parametersof Nonlinear Regression withNonlinearInequalityConstraints................................. 10 1.2.1 StatementoftheProblemandaMethodofItsSolution...... 10 1.2.2 SolutiontotheAuxiliaryProblem............................. 19 1.2.3 CompatibilityofConstraintsintheAuxiliaryProblem....... 19 1.2.4 CalculationoftheConstants(cid:2) andı.......................... 24 1.3 EstimationofMultivariateLinearRegressionParameters withNonlinearEqualityConstraints................................... 25 2 Asymptotic Properties of Parameters in Nonlinear RegressionModels........................................................... 29 2.1 ConsistencyofEstimatesinNonlinearRegressionModels........... 29 2.2 Asymptotic Properties of Nonlinear Regression ParametersEstimatesObtainedbytheLeastSquares MethodUnderaPrioryInequalityConstraints(ConvexCase)....... 38 2.2.1 Introduction..................................................... 38 2.2.2 AuxiliaryResults ............................................... 40 2.2.3 FundamentalResults ........................................... 52 2.3 Asymptotic Properties of Nonlinear Regression ParametersEstimates bythe LeastSquaresMethod UnderaPrioryInequalityConstraints(Non-ConvexCase)........... 57 2.3.1 AssumptionsandAuxiliaryResults ........................... 57 2.3.2 FundamentalResult............................................. 58 ix x Contents 2.4 Limit Distribution of the Estimate of Regression ParametersWhichAreSubjecttoEqualityConstraints............... 61 2.5 AsymptoticPropertiesoftheLeastSquaresEstimates ofParametersofaLinearRegressionwithNon-Stationary VariablesUnderConvexRestrictionsonParameters.................. 64 2.5.1 Settings.......................................................... 64 2.5.2 ConsistencyofEstimator....................................... 65 2.5.3 LimitDistributionoftheParameterEstimate ................. 67 3 Method of Empirical Means in Nonlinear Regression andStochasticOptimizationModels ...................................... 73 3.1 Consistency of Estimates Obtainedby the Method of Empirical Means with IndependentOr Weakly DependentObservations................................................ 74 3.2 RegressionModelsforLongMemorySystems ....................... 81 3.3 Statistical Methods in Stochastic Optimization andEstimationProblems ............................................... 85 3.4 EmpiricalMeanEstimatesAsymptoticDistribution.................. 89 3.4.1 AsymptoticDistributionofEmpiricalEstimates for Models with Independentand Weakly DependentObservations........................................ 89 3.4.2 AsymptoticDistributionofEstimatesforLong MemoryStochasticSystems ................................... 99 3.4.3 AsymptoticDistributionoftheLeastSquares EstimatesforLongMemoryStochasticSystems ............. 101 3.5 LargeDeviationsofEmpiricalMeansin Estimation andOptimizationProblems............................................. 104 3.5.1 LargeDeviationsoftheEmpiricalMeansMethod forDependentObservations.................................... 104 3.5.2 Large Deviations of Empiric Estimates forNon-StationaryObservations............................... 112 3.5.3 LargeDeviationsinNonlinearRegressionProblems......... 118 4 DeterminationofAccuracyofEstimationofRegression ParametersUnderInequalityConstraints................................ 121 4.1 PreliminaryAnalysisoftheProblem................................... 121 4.2 Accuracy of Estimation of Nonlinear Regression Parameters:TruncatedEstimates....................................... 123 4.3 DeterminationoftheTruncatedSampleMatrixofm.s.e. oftheEstimateofParametersinNonlinearRegression............... 137 4.4 AccuracyofParameterEstimationinLinearRegression withConstraintsandwithoutaTrend.................................. 138 4.4.1 AuxiliaryResults ............................................... 139 4.4.2 MainResults.................................................... 148 4.5 Determinationof AccuracyofEstimation ofLinear RegressionParametersinRegressionwithTrend ..................... 154

See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.